newsletter-icon

Get the latest updates in your inbox

Subscribe
Toxicity Report #3
Toxicity Report #3

Toxicity Report #3

It's been a while since the last time we spoke about Predecessor's anti-toxicity measures so today we're sharing a little refresher about how exactly our system works!

Hello,

Welcome to the third instalment of the Predecessor toxicity report!

For the first time, we’re saying hello to players on multiple platforms, as our PlayStation beta is currently ongoing! If you haven’t had a chance to try Predecessor on console, don’t miss your chance to check it out here.

Because it has been a while since the last time we spoke about anti-toxicity measures, we appreciate that some of you might be confused about how exactly our system works, so below we're including a little refresher.

When setting out to tackle issues of toxicity, we have a three-pronged approach utilizing a mixture of automatic detection, AI chat moderation, and manual actioning.

How we tackle toxicity

Automatic Detection

The first of our toxicity-mitigation measures is automated AFK and Draft Disconnection detection.  We use an escalating series of penalties for both AFK and Draft Disconnections, allowing us to be lenient for unexpected disconnections or life events, but with increasingly severe matchmaking bans for repeat offences. The goal of this system is to be a first line of defense to catch cases where a player was clearly and obviously contributing to a negative experience, either in draft or during a match.

Report Action

In addition to automatic detection, we also rely on the reports you make during matches to catch bad actors.  When a player is intentionally feeding the other team, is harassing other players in chat, or is otherwise making a clear effort to ruin the experience for everyone, please use the report feature! We collect and run your reports through a tool that allows us to identify and issue bans for the players that have clearly crossed the line. 

Of course, we are careful about identifying the difference between a rough game and malicious trolling when handing out bans, and thus some may feel our punishments are too light at times. We hear your feedback, and we’re continuously evaluating report data as we try to find the best balance between shutting down trolls and troublemakers while sparing those that are otherwise doing their best but might have still been reported. We’re also planning to implement a “Thanks for Reporting!” feedback loop in the future to let you know when a player you’ve reported has been suspended, but that requires some custom tech that’s not available with the current tools we have. Soon™!

Chat Moderation

The third pillar of our Toxicity Triangle is automatic chat moderation. To help keep chat civil, we use an AI-based tool called GGWP. GGWP is a natural language processing model that we’ve had specifically trained for the Predecessor game and community. Rather than feeding a list of bad words to a bot that would sanction everyone equally regardless of context and prior chat history, the AI actually assesses several things, including:

  • The vocabulary used in the messages across 16 languages (Arabic, Chinese, English, French, German, Indonesian, Italian, Japanese, Korean, Portuguese, Polish, Russian, Spanish, Thai, Turkish and Vietnamese)

  • The tone of the conversation taking place

  • The chat history (both recent and extended) of all conversing members

This allows us to take a more nuanced approach, allowing for swearing or a bit of a salty rant in the appropriate context, but with a focus on limiting targeted toxicity and harassment among players. This also means two players saying the same thing might see different results depending on context, reputation, and other circumstances fed into the AI. If a player only ever uses chat to trash talk others, the AI will be less merciful and penalties will escalate up to and including permanent chat suspension.

What difference does it make?

97% of all players who have ever used in-game text chat have a “moderate” or better chat history, and only 9% of players have ever received any kind of persistent mute sanction. Those numbers may seem surprisingly good in a competitive MOBA, but in reality we’ve found that a few bad apples tend to have a really outsized influence on the tone and toxicity level of a match, which is why a tool like GGWP that allows for message moderation as it happens is so important.

There are exceptions, of course. Derogative and very offensive language like slurs will get you silenced immediately regardless of how good your reputation is. And it doesn’t matter who was being addressed, whether it was a real person or a character from the game - that sort of language has no place in Predecessor, period.

We get that everyone is only human. It’s acceptable to get a little worked up if a game isn’t going your way or you’re just not on the same page as your teammates sometimes, but everything that actually needs to be communicated in the game can be said without flaming your teammates or resorting to targeted or hurtful language.

How do I steer clear of the AI?

  • Use chat to communicate strategy and banter with other players - if you’re not in the mood to be nice, don’t say anything.

  • If someone is being toxic, report them and mute them. Don’t be toxic back at them; that’s not a solution and you’ll both end up sanctioned.

  • Compliment your teammates when they do something awesome. Being kind to others will go a long way.

  • The AI will notice if you try to d0dg3 w0rd f1lt3rs to say something nasty, and you are subject to the same sanctions regardless of your creative spelling or w o r d  s p a c i n g. We regularly update our systems to include an increasingly bigger database of such creations.

  • You can always disable the in-game chat in the Game Settings if you don’t want to participate in the conversation for whatever reason.

Feedback and Improvement

Mistaken multi-match suspensions continue to be extremely rare. For example, only about 2% of the longer chat suspensions we’ve checked were applied incorrectly.  The model does use a lower threshold (with a slightly lower accuracy) for a single-match mute to allow for quicker shutdown of potentially toxic messages, but those short mutes are not part of the longer escalating suspensions and do not have any associated account penalties.

Many of the adjustments we’ve made to the system since its launch have been because of the messages and potential corrections you’ve flagged for us. Your reports are a tremendous help as we continue to work with GGWP to refine the AI model over time, so keep them coming!  If you have any issues, feedback, or other suggestions, you can always reach out on Discord to members of the Omeda Community team (particularly yours truly, Omeda_Zuzu) or via a support ticket.

Finally, we can confirm that the much-requested feature, the ability to mute pings, is expected to arrive in V0.15! This should hopefully further improve your experience with the game.

- Omeda Zuzu (Community Experience Manager)
- JShredz (Senior Game Ops Manager)